Facial Image Obfuscation Training for Privacy-Reconstruction Balance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image obfuscation techniques struggle to balance the degree of obfuscation with the ability to reconstruct identification information from obfuscated facial images, leading to either high obfuscation with low reconstruction or low obfuscation with high reconstruction.
Innovation Solution
A neural network is trained using a backpropagation refinement scheme to perform averaging, warping, and noise addition transformations on facial images, while extracting features to achieve a trade-off between obfuscation and reconstruction through parameter optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If image obfuscation techniques are applied to conceal facial information, then human indecipherability is improved, but machine decipherability deteriorates
Solution Approach 1:
The patent applies different processing treatments to different regions of the facial image. Specifically, it identifies key facial regions (eyes, nose, mouth) and applies selective obfuscation to non-key regions while preserving critical features in key regions. This local differentiation allows the system to achieve high human indecipherability for non-identified regions while maintaining machine decipherability for identification purposes.
Solution Approach 2:
The patent dynamically adjusts obfuscation parameters such as noise intensity, blur level, and transformation strength based on the importance of different facial regions. By changing these parameters locally across the image, the system optimizes the balance between making the image unrecognizable to humans while preserving sufficient information for machine recognition algorithms.
2Reliability
If strong obfuscation transformations are applied to facial images, then privacy protection is improved, but reconstruction accuracy deteriorates
Solution Approach 1:
The patent performs preliminary identification of key facial regions before applying obfuscation transformations. By pre-marking critical areas that must be preserved for accurate reconstruction, the system can apply stronger obfuscation to non-key regions while ensuring that regions necessary for reconstruction maintain sufficient quality. This preliminary action enables the system to achieve both strong privacy protection and accurate reconstruction.
Solution Approach 2:
The patent employs a feedback mechanism where the obfuscation process is iteratively applied and evaluated. After each obfuscation pass, the system evaluates both privacy protection effectiveness and reconstruction quality, then adjusts the obfuscation strength accordingly. This feedback loop allows the system to find the optimal balance point where privacy is sufficiently protected while reconstruction accuracy is maintained.
Data Source
AI summary
A method of training a neural network configured to obfuscate a facial image and an electronic device for performing the method are provided. The method includes obtaining, based on an input facial image, an output facial image in which the input facial image is obfuscated, extracting, based on the input facial image, a feature of the input facial image for reconstructing identification information included in the input facial image from the output facial image, extracting, based on the output facial image, a feature of the output facial image corresponding to the feature of the input facial image, and training the neural network based on a difference between the feature of the input facial image and the feature of the output facial image.